Update README.md
Browse files
README.md
CHANGED
|
@@ -19,11 +19,12 @@ This is a [Whisper large](https://huggingface.co/openai/whisper-large-v2) finetu
|
|
| 19 |
Training was done using Huggingface and Deepspeed with ZeRO stage 2.
|
| 20 |
|
| 21 |
* learning rate: 1e-5
|
| 22 |
-
* optimizer: CPUAdamW (
|
| 23 |
* lr scheduler: linear
|
| 24 |
* warmup steps: 500
|
| 25 |
* per device batch size: 20
|
| 26 |
* GPUs: 8 x NVIDIA A100 40GB
|
| 27 |
* total batch size: 160
|
| 28 |
* steps: 20000
|
|
|
|
| 29 |
* entire decoder was frozen
|
|
|
|
| 19 |
Training was done using Huggingface and Deepspeed with ZeRO stage 2.
|
| 20 |
|
| 21 |
* learning rate: 1e-5
|
| 22 |
+
* optimizer: CPUAdamW (Deepspeed)
|
| 23 |
* lr scheduler: linear
|
| 24 |
* warmup steps: 500
|
| 25 |
* per device batch size: 20
|
| 26 |
* GPUs: 8 x NVIDIA A100 40GB
|
| 27 |
* total batch size: 160
|
| 28 |
* steps: 20000
|
| 29 |
+
* fp16
|
| 30 |
* entire decoder was frozen
|